Motivation: Who Cares About I / O? CS252 Graduate Computer Architecture • CPU Perf ormance: 60% per year • I / O syst em perf ormance limit ed by mechanical delays (disk I / O) I / O I ntroduction: Storage Devices & RAI D < 10% per year (I O per sec) • Amdahl' s Law: syst em speed- up limited by the slowest part! 10% I O & 10x CPU => 5x Perf ormance (lose 50%) Jason Hill 10% I O & 100x CPU => 10x Perf ormance (lose 90%) • I / O bottleneck: Diminishing f raction of time in CPU Diminishing value of f aster CPUs CS252/ Culler CS252/ Culler 2/ 7/ 02 2/ 7/ 02 Lec 6. 1 Lec 6. 2 I / O Systems Big Picture: Who cares about CPUs? interrupts interrupts Processor • Why still important to keep CPUs busy vs. I O devices ("CPU t ime"), as CPUs not cost ly? Cache – Moore' s Law leads to both large, f ast CPUs but also to very small, cheap CPUs – 2001 Hypothesis: 600 MHz PC is f ast enough f or Of f ice Memory - I/O Bus Tools? – PC slowdown since f ast enough unless games, new apps? Main I/O I/O I/O • People care more about about st oring inf ormat ion Memory Controller Controller Controller and communicat ing inf ormat ion t han calculat ing – "I nf ormation Technology" vs. "Computer Science" Graphics Disk Disk Network – 1960s and 1980s: Computing Revolution – 1990s and 2000s: I nf ormation Age • Next 3 weeks on st orage and communicat ion CS252/ Culler CS252/ Culler 2/ 7/ 02 2/ 7/ 02 Lec 6. 3 Lec 6. 4 Storage Technology Drivers Outline • Driven by t he prevailing comput ing paradigm • Disk Basics – 1950s: migration f rom batch to on- line processing • Disk History – 1990s: migration to ubiquitous computing • Disk opt ions in 2000 » computers in phones, books, cars, video cameras, … • Disk f allacies and perf ormance » nationwide f iber optical network with wireless tails • FLASH • Ef f ect s on st orage indust ry: – Embedded storage • Tapes » smaller, cheaper, more reliable, lower power • RAI D – Data utilities » high capacity, hierarchically managed storage CS252/ Culler CS252/ Culler 2/ 7/ 02 2/ 7/ 02 Lec 6. 5 Lec 6. 6 P age 1
Photo of Disk Head, Arm, Disk Device Terminology Actuator Inner Outer Arm Head Sector Spindle Track Track Arm Head Platter Actuator { Actuator • Several platters, with inf ormation recorded magnetically on both surfaces (usually) Platters (12) • Bits recorded in tracks, which in turn divided into sectors (e.g., 512 Bytes) • Actuator moves head (end of arm,1/ surf ace) over track (“seek” ), select surf ace, wait f or sector rotate under head, then read or write – “Cylinder”: all tracks under heads CS252/ Culler CS252/ Culler 2/ 7/ 02 2/ 7/ 02 Lec 6. 7 Lec 6. 8 Disk Device Perf ormance Disk Device Perf ormance Inner Outer Sector Head Controller Arm • Average distance sector f rom head? Spindle Track Track • 1/ 2 t ime of a rot at ion – 10000 Revolutions Per Minute ⇒ 166.67 Rev/ sec – 1 revolution = 1/ 166.67 sec ⇒ 6.00 milliseconds Platter Actuator – 1/ 2 rotation (revolution) ⇒ 3.00 m s • Average no. t racks move arm? • Disk Lat ency = Seek Time + Rot at ion Time + Transf er – Sum all possible seek distances Time + Cont roller Overhead f rom all possible tracks / # possible • Seek Time? depends no. tracks move arm, seek speed of disk » Assumes average seek distance is random • Rotation Time? depends on speed disk rotates, how f ar sector is – Disk industry standard benchmark f rom head • Transf er Time? depends on data rate (bandwidth) of disk (bit density), size of request CS252/ Culler CS252/ Culler 2/ 7/ 02 2/ 7/ 02 Lec 6. 9 Lec 6. 10 Data Rate: I nner vs. Outer Tracks Devices: Magnetic Disks • To keep t hings simple, orginally kept same number of • Purpose: Track sect ors per t rack – Long- term, nonvolatile storage Sector – Since outer track longer, lower bits per inch – Large, inexpensive, slow level in the storage hierarchy • Compet it ion ⇒ decided to keep BPI the same f or all • Characteristics: tracks (“const ant bit densit y”) Cylinder ⇒ More capacity per disk – Seek Time (~8 ms avg) Platter ⇒ More of sectors per track towards edge » positional latency Head ⇒ Since disk spins at constant speed, » rotational latency outer tracks have f aster data rate Transf er rate • 7200 RPM = 120 RPS => 8 ms per rev ave rot. latency = 4 ms • Bandwidt h out er t rack 1. 7X inner t rack! – 10- 40 MByte/ sec 128 sectors per track => 0.25 ms per sector 1 KB per sector => 16 MB / s – I nner track highest density, outer track lowest, so not really – Blocks const ant • Capacit y – 2. 1X length of track outer / inner, 1. 7X bits outer / inner – Gigabytes Response time – Quadruples every 2 years = Queue + Controller + Seek + Rot + Xfer (aerodynamics) Service time CS252/ Culler CS252/ Culler 2/ 7/ 02 2/ 7/ 02 Lec 6. 11 Lec 6. 12 P age 2
Disk Perf ormance Model / Trends State of the Art: Barracuda 180 – 181. 6 GB, 3. 5 inch disk • Capacit y – 12 plat t ers, 24 surf aces + 100%/ year (2X / 1.0 yrs) – 24, 247 cylinders • Transf er rate (BW) Track + 40%/ year (2X / 2.0 yrs) – 7, 200 RPM; (4. 2 ms avg. lat ency) • Rotation + Seek time – 8%/ year (1/ 2 in 10 yrs) Sector – 7. 4/ 8. 2 ms avg. seek (r/ w) • MB/ $ Cylinder > 100%/ year (2X / 1.0 yrs) – 64 t o 35 MB/ s (int ernal) Track Arm Platter Head Fewer chips + areal density Buffer – 0. 1 ms controller time Latency = – 10. 3 wat t s (idle) Queuing Time + per byte { Controller time + per access Seek Time + + Rotation Time + Size / Bandwidth source: www.seagate.com CS252/ Culler CS252/ Culler 2/ 7/ 02 2/ 7/ 02 Lec 6. 13 Lec 6. 14 Disk Perf ormance Example (will f ix later) CS 252 Administrivia • Calculat e t ime t o read 64 KB (128 sect ors) f or Barracuda 180 X using advert ised perf ormance; sector is on outer track Disk lat ency = average seek t ime + average rot at ional delay + t ransf er t ime + cont roller overhead = 7. 4 ms + 0. 5 * 1/ (7200 RPM) + 64 KB / (65 MB/ s) + 0. 1 ms = 7. 4 ms + 0. 5 / (7200 RPM/ (60000ms/ M)) + 64 KB / (65 KB/ ms) + 0. 1 ms = 7. 4 + 4. 2 + 1. 0 + 0. 1 ms = 12. 7 ms CS252/ Culler CS252/ Culler 2/ 7/ 02 2/ 7/ 02 Lec 6. 15 Lec 6. 16 Areal Density Areal Density Year Areal Density • Bits recorded along a track 100000 1973 1.7 – Metric is Bits Per I nch (BPI ) 1979 7.7 10000 1989 63 • Number of t racks per surf ace 1997 3090 1000 2000 17100 – Metric is Tracks Per I nch (TPI ) Areal Density 100 • Disk Designs Brag about bit densit y per unit area 10 – Metric is Bit s Per Square I nch – Called Areal Density 1 – Areal Density = BPI x TPI 1970 1980 1990 2000 Year – Areal Density = BPI x TPI – Change slope 30%/ yr to 60%/ yr about 1991 CS252/ Culler CS252/ Culler 2/ 7/ 02 2/ 7/ 02 Lec 6. 17 Lec 6. 18 P age 3
Historical Perspective MBits per square inch: • 1956 I BM Ramac — early 1970s Winchester DRAM as % of Disk over time – Developed f or mainf rame computers, proprietary interf aces – Steady shrink in f orm f actor: 27 in. to 14 in 9 v. 22 Mb/si 50% • Form f act or and capacit y drives market , more t han perf ormance 40% • 1970s: Mainf rames ⇒ 14 inch diamet er disks • 1980s: Minicomput ers, Servers ⇒ 8”, 5 1/ 4 ” diameter 30% • PCs, workst at ions Lat e 1980s/ Early 1990s: 20% 470 v. 3000 Mb/si – Mass market disk drives become a reality 10% » industry standards: SCSI , I PI , I DE Cs ⇒ 3. 5 inch diameter disks – Pizzabox P 0.2 v. 1.7 Mb/si 0% – Laptops, notebooks ⇒ 2. 5 inch disks – Palmtops didn’t use disks, 1974 1980 1986 1992 1998 2000 so 1. 8 inch diameter disks didn’t make it • 2000s: source: New York Times, 2/23/98, page C3, “Makers of disk drives crowd even mroe data into even smaller spaces” – 1 inch f or cameras, cell phones? CS252/ Culler CS252/ Culler 2/ 7/ 02 2/ 7/ 02 Lec 6. 19 Lec 6. 20 Disk History Disk History Data density Mbit/sq. in. Capacity of Unit Shown Megabytes 1989: 1997: 1997: 63 Mbit/sq. in 1450 Mbit/sq. in 3090 Mbit/sq. in 1973: 1979: 60,000 MBytes 2300 MBytes 8100 MBytes 1. 7 Mbit/sq. in 7. 7 Mbit/sq. in 140 MBytes 2,300 MBytes source: New York Times, 2/23/98, page C3, source: New York Times, 2/23/98, page C3, “Makers of disk drives crowd even more data into even smaller spaces” “Makers of disk drives crowd even mroe data into even smaller spaces” CS252/ Culler CS252/ Culler 2/ 7/ 02 2/ 7/ 02 Lec 6. 21 Lec 6. 22 1 inch disk drive! Disk Characteristics in 2000 • 2000 I BM MicroDrive: Seagate IBM IBM 1GB – 1.7” x 1.4” x 0.2” Cheetah Travelstar Microdrive – 1 GB, 3600 RPM, ST173404LC 32GH DJSA - DSCM-11000 5 MB/ s, 15 ms seek Ultra160 SCSI 232 ATA-4 – Digital camera, PalmPC? Disk diameter 3.5 2.5 1.0 • 2006 MicroDrive? (inches) Formatted data 73.4 32.0 1.0 • 9 GB, 50 MB/ s! capacity (GB) Cylinders – Assuming it f inds a niche 14,100 21,664 7,167 in a successf ul product Disks – Assuming past trends continue 12 4 1 Recording 24 8 2 Surfaces (Heads) Bytes per sector 512 to 4096 512 512 Avg Sectors per ~ 424 ~ 360 ~ 140 track (512 byte) Max. areal 6.0 14.0 15.2 density(Gbit/sq.in.) CS252/ Culler $ 828 $ 447 $ 435 CS252/ Culler 2/ 7/ 02 2/ 7/ 02 Lec 6. 23 Lec 6. 24 P age 4
Recommend
More recommend
